Human Leukocyte Antigen (HLA) B27 Allotype-Specific Binding and Candidate Arthritogenic Peptides Revealed through Heuristic Clustering of Data-independent Acquisition Mass Spectrometry (DIA-MS) Data

Abstract

Expression of HLA-B27 is strongly associated with ankylosing spondylitis (AS) and other spondyloarthropathies. While this is true for the majority of HLA-B27 allotypes, HLA-B*27:06 and HLA-B*27:09 are not associated with AS. These two subtypes contain polymorphisms that are ideally positioned to influence the bound peptide repertoire. The existence of disease-inducing peptides (so-called arthritogenic peptides) has therefore been proposed that are exclusively presented by disease-associated HLA-B27 allotypes. However, we have recently demonstrated that this segregation of allotype-bound peptides is not the case and that many peptides that display sequence features predicted to favor binding to disease-associated subtypes are also capable of being presented naturally by protective alleles. To further probe more subtle quantitative changes in peptide presentation, we have used a combination of data-independent acquisition (DIA) and multiple reaction monitoring (MRM) mass spectrometry to quantify the abundance of 1646 HLA-B27 restricted peptides across the eight most frequent HLA-B27 allotypes (HLA-B*27:02-HLA-B*27:09). We utilized K means cluster analysis to group peptides with similar allelic binding preferences across the eight HLA-B27 allotypes, which enabled us to identify the most-stringent binding characteristics for each HLA-B27 allotype and further refined their existing consensus-binding motifs. Moreover, a thorough analysis of this quantitative dataset led to the identification of 26 peptides, which are presented in lower abundance by HLA-B*27:06 and HLA-B*27:09 compared with disease-associated HLA-B27 subtypes. Although these differences were observed to be very subtle, these 26 peptides might encompass the sought-after arthritogenic peptide(s).

Human leukocyte antigen (HLA) class I molecules present endogenous peptides derived from self-proteins or intracellular pathogens at the cell surface for scrutiny by CD8⁺ T lymphocytes. These peptide ligands are constrained within the peptide-binding cleft, which consists of six binding pockets (A-F) that accommodate the amino acid side chains of the peptides (1). Polymorphisms characterizing and differentiating HLA alleles and allotypes mostly cluster around these binding pockets, thereby shaping and defining their specific peptide cargo.

HLA-B27, which belongs to the family of HLA class I molecules, is an extremely polymorphic B-locus with more than 100 allelic variants known to date (2, 3). The most frequent allotypes with comprehensively described peptide repertoires are HLA-B*27:02–HLA-B*27:09 (4–18). Due to the architecture of their B pocket, the most dominant feature of all HLA-B27 bound peptides is the presence of an Arg residue at position P2 (19–21). Additional primary and secondary anchor positions are P3, PΩ-2, and PΩ, which mostly carry hydrophobic and aromatic amino acid residues that bind into the D, E, and F pockets, respectively (18, 20, 21). All other peptide positions are considered as nonanchor positions and tolerate a large variety of amino acid residues with notable differences between the various allotypes (summarized in (18).

Ankylosing spondylitis (AS)¹, a chronic inflammatory autoimmune disease, and other spondyloarthropathies are stronglylinked to the expression of HLA-B27 (reviewed in (22, 23). Even though this association has been known for more than 40 years (24, 25), the pathogenic property of HLA-B27 remains still unknown and many hypotheses are currently being investigated to elucidate its apparent role in AS (reviewed in (26, 27). Due to the peptide-presenting function of HLA-B27 and other HLA molecules, one of the most popular hypotheses is the arthritogenic peptide theory. This theory assumes that cytotoxic T lymphocytes (CTL) primed against pathogenic HLA-B27-restricted antigen(s) cross-react on molecularly similar self-peptide(s), which subsequently leads to tissue damage and onset of autoimmunity. In support of this theory, onset of AS is often preceded by infection with enteric bacteria and HLA-B27-restricted CTLs have been identified in the synovial fluid of AS patients, which recognize both bacterial epitopes and self-peptides (28, 29). Moreover, two allelic variants of HLA-B27 (HLA-B*27:06 and HLA-B*27:09), which contain polymorphisms ideally placed to alter the presented peptide repertoire, are not associated with AS (30–34).

Based on this apparent protective role of HLA-B*27:06 and HLA-B*27:09 in disease pathogenesis, a number of analyses focused on comparing the peptide repertoires of disease-associated and nonassociated HLA-B27 subtypes with the goal to identify putative arthritogenic peptides, which were expected to be qualitatively absent from nondisease-associated subtypes (11–14, 17, 18). However, such a peptide has not been described yet, and a peptide feature that stratifies disease- and nondisease-associated subtypes does not appear to exist. In agreement of this notion, we could recently show that a peptide, which is capable of binding to each disease-associated HLA-B27 subtype, will also bind to the nondisease-associated allotypes HLA-B*27:06 and HLA-B*27:09 (18). This observation challenged the arthritogenic peptide theory in such that absolute binding preferences do not explain disease association. However, quantitative changes in peptide presentation between disease-associated and nonassociated subtypes may be relevant for disease pathogenesis by exceeding (or falling below) a threshold required for the activation of autoreactive T cells and, thus, for the onset of autoimmunity. Therefore, we sought to quantify the abundance of known HLA-B27 specific peptides across the eight most frequent HLA-B27 allotypes (HLA-B*27:02–HLA-B*27:09) with the goal to specifically identify candidate arthritogenic peptides that are expected to be presented in lower abundance by HLA-B*27:06 and HLA-B*27:09 compared with disease-associated HLA-B27 subtypes.

Data-independent acquisition (DIA) mass spectrometry—often referred to as SWATH-MS on SCIEX TripleTOF^® instruments—is the state-of-the-art method to quantify thousands of peptides across multiple samples (35, 36). A mass spectrometer operated in DIA/SWATH mode uses wide ms1 isolation windows (typically 20–25 Da) to scan sequentially and repeatedly through a given mass range. All precursor ions isolated within each window are simultaneously fragmented in the collision cell resulting in multipeptide-derived ms2 spectra. Even though these spectra are too complex to be interrogated by conventional search engine algorithms, the continuous acquisition of all fragment ions over the chromatographic run time allows the extraction of quantifiable ms2 traces for virtually every peptide of interest. However, a faithful and accurate evaluation of DIA/SWATH-MS data requires prior knowledge of spectral and chromatographic information for all peptides of interest, although significant efforts are currently being made to develop software packages that extract sequence information directly from DIA/SWATH data (for example DIA Umpire: (37)).

Despite unique and compelling advantages of DIA/SWATH technology, multiple reaction monitoring (MRM)-MS still excels in unparalleled sensitivity (38, 39). This level of sensitivity is achieved by analyzing only one fragment ion (transition) at a time on a triple quadrupole mass spectrometer, which combines excellent ion transmission efficiencies with dual ion selectivity processes to significantly reduce background noise in favor of the analyte of interest. However, MRM-MS is severely restricted by a relatively low number of peptides that can be analyzed in one acquisition.

In this study, we used a combination of SWATH- and MRM-MS to quantify the abundance of 1646 epitopes across eight different HLA-B27 allotypes (HLA-B*27:02-HLA–B*27:09), including the non-AS associated subtypes HLA-B*27:06 and HLA-B*27:09. Computational algorithms were utilized to group all quantified peptides into distinct clusters with similar binding preferences across the HLA-B27 allotypes investigated. A detailed analysis of these cluster-specific consensus-binding motifs provided a means to attain the most stringent binding characteristics for each HLA-B27 allotype and further refined our understanding of the intricate mechanisms underlying HLA-peptide binding. Of note, a thorough comparison of this comprehensive quantitative dataset revealed 26 putative arthritogenic peptides, whose immunological potential will be analyzed in subsequent studies.

EXPERIMENTAL PROCEDURES

Experimental Design and Statistical Rationale

We recently described the comprehensive analysis of the peptide repertoires of the eight most frequent HLA-B27 allotypes (HLA-B*27:02–HLA-B*27:09) by LC-MS/MS (18). In brief, HLA-B27 bound peptides were isolated from stable C1R transfectants by immunoaffinity capture, separated by reverse-phase high-performance liquid chromatography (RP-HPLC), and sequenced on a 5600⁺ TripleTOF^® (SCIEX, Framingham, USA) mass spectrometer. At least four independent immunoprecipitations (biological replicates) were conducted to establish comprehensive immunopeptidomes for each HLA-B27 allomorph.

To obtain sufficient material for the eight SWATH-MS and 152 MRM-MS acquisitions presented in this study, the remaining aliquots of these subtype-specific peptide eluates were combined to form one biological pool for each of the eight distinct samples (HLA-B*27:02–HLA-B*27:09). By its nature, SWATH-MS eliminates technical variation allowing comprehensive interrogation of such pooled biological samples.

Data-Independent Acquisition (SWATH-MS)

Samples were loaded for 15 min with an isocratic flow (5 μl/min) of loading buffer (0.1% formic acid, 2% acetonitrile) onto a trap column (Eksigent, Dublin, USA; 200 μm × 0.5 mm ChromXP C18-CL 3 μm 120Å) using an Eksigent NanoUltra cHiPLC system. A microfluidic analytical column (Eksigent; 75 μm × 15 cm ChromXP C18-CL 3 μm 120Å) was switched in-line to separate the peptides by increasing concentrations of 80% acetonitrile /0.1% formic acid at a flow of 300 nl/min for 85 min.

The peptides were analyzed on a 5600⁺ TripleTOF mass spectrometer (SCIEX) operated in positive ion and SWATH-MS mode. In each cycle, a 200 ms precursor scan (range: 300–1800 Da) preceded the acquisition of 28 sequential SWATH-MS scans with an accumulation time of 100 ms each resulting in a total cycle time of 3 s. Each of the 28 SWATH-MS acquisitions scanned a 25 Da window between 300 to 1000 Da.

Raw SWATH data files have been deposited at PeptideAtlas (www.peptideatlas.org) with the dataset identifier PASS00763 (password: DT4743iu).

SWATH-MS data analysis

All SWATH-MS data were evaluated with the open source software Skyline (v2.6.0.7176; (40). The six most abundant b- or y- fragment ions of each peptide were selected for monitoring based on a spectral library, which contained information of all HLA-B27 peptides identified in our previous study (18). A custom-made retention time predictor based on retention time alignment of 15 intrinsic HLA-B27 peptides was used to correct retention time drift between analyses (Supplemental Fig. 1).

SWATH-MS data were imported into Skyline and the performance of the peak picking was visually assessed for each peptide and, if necessary, manually adjusted. A peptide identification was considered successful if all of the following criteria were met in at least one sample: (i) the deviation between the peptide's observed and predicted retention time was less than 5 min, (ii) the dot product, which delineates the similarity between the library spectrum and the observed transition chromatograms, was greater than 0.8, (iii) the mass error was smaller than 5 ppm, and (iv) the maximal peak height of a transition exceeded 500 arbitrary intensity units, which prevents quantification of low abundant background peaks.

MRM-MS Acquisition

A QTRAP^® 5500 mass spectrometer (SCIEX) coupled to an Eksigent NanoUltra cHiPLC system was used for MRM detection as described previously (41, 42). In brief, the sample was loaded via a microfluidic trap column (Eksigent; 200 μm × 0.5 mm ChromXP C18-CL 3 μm 120Å) onto a microfluidic analytical column (Eksigent; 75 μm × 15 cm ChromXP C18-CL 3 μm 120Å), and the peptides were separated by increasing concentrations of 98% acetonitrile/0.1% formic acid at a flow rate of 300 nl/min for 90 min. The QTRAP^® 5500 was operated in MRM mode (unit resolution for Q1 and Q3) analyzing each transition for 5 ms per cycle (dwell time). After each cycle, a single enhanced product ion scan (10,000 Da/s; rolling collision energy; unit resolution) was triggered to obtain sequence information of the most abundant precursor ion during this cycle.

Raw MRM data files have been deposited at PeptideAtlas (www.peptideatlas.org) with the dataset identifier PASS00762 (password: CS3537k).

MRM-MS Data Analysis

MRM data analysis and peptide quantification was done in Skyline (v2.6.0.7176; (40) using identical evaluation criteria as described for the SWATH-MS data analysis with the exception of the mass error criterion, which could not be used due to the modality of MRM-MS. Of note, instead of measuring the peptides of the iRT-B27 retention time predictor in each MRM-MS run, synthetic versions (GL Biochem, Shanghai, China) were separately analyzed in technical quadruplicates to determine the MRM-gradient-specific regression curve of the iRT-B27 peptides. The equation of this regression curve was applied to all MRM-MS runs for retention time alignment (Supplemental Fig. 2).

Data Normalization

To correct for differences in sample density and loading, the peak areas for each peptide were normalized based on the assumption that a peptide with a universal HLA-B27 consensus-binding motif is presented in equal amounts across the eight HLA-B27 allotypes investigated. To minimize the intrinsic error of this assumption, individual normalization values were calculated for 428 universal HLA-B27 peptides and subsequently averaged (Supplemental Fig. 3; Supplemental Table 1). A peptide was considered a universal HLA-B27 peptide if it (i) was 9–12 amino acids in length, (ii) contained an Arg residue at P2, (iii) ended in a Phe, Leu, Ile, or Val residue, and (iv) did not carry a His, Arg, or Lys residue at P1 (18).

Cluster Analysis

The k-means algorithm of the TM4:MeV software suite (http://www.tm4.org/mev.html; (43) was used to group the peptides according to their relative expression levels into 30 distinct clusters using the default parameters.

RESULTS

Quantification of B27-Specific Peptides across the Eight Most Frequent HLA-B27 Allotypes

We utilized a combination of SWATH-MS, a DIA method accessible on TripleTOF^® 5600⁺ instruments (SCIEX), and MRM-MS to comprehensively and accurately quantify HLA-B27-specific peptides across the eight most frequent HLA-B27 allotypes (HLA-B*27:02–HLA-B*27:09) including the nondisease-associated subtypes HLA-B*27:06 and HLA-B*27:09 (see Fig. 1 for experimental workflow). Equal amounts of subtype-specific HLA-B27 peptide eluates were pooled and analyzed on a TripleTOF^® 5600⁺ mass spectrometer (SCIEX) operated in SWATH-MS mode. The data were evaluated in Skyline (40) based on a comprehensive library containing spectral information of peptides isolated from all HLA-B27 allotypes and a custom-made retention time predictor to align peak picking across each sample (iRT-B27 retention time predictor; Supplemental Fig. 1). In total, we analyzed 3306 peptides encompassing 3277 high confidence HLA-B27-bound peptides from our previous studies (18) and 29 selected HLA-Cw4-specific peptides, which are present in all the eluates due to low level expression of HLA-Cw4 on the parental C1R cell line (18, 44); Supplemental Table 2).

Fig. 1.

Workflow. We recently determined the peptide repertoires of the eight most frequent HLA-B27 allotypes (HLA-B*27:02 - HLA-B*27:09) by analyzing 44 independent DDA acquisitions on a TripleTOF 5600⁺ (SCIEX) mass spectrometer (18). Sequence information was obtained using the Paragon algorithm implemented in the ProteinPilot 4.5 software suite (SCIEX) and all ProteinPilot search result (.group) files were used to generate a spectral library in Skyline (MacLean et al. 2010). A custom-made retention time predictor (iRT-B27 retention time predictor) consisting of 15 intrinsic HLA-B27 and HLA-Cw4 peptides was created and used for retention time alignment (Supplemental Fig. 1). Residual, HLA-B27 allotype specific samples were pooled and analyzed by DIA/SWATH-MS on a TripleTOF 5600⁺ (SCIEX) mass spectrometer. Data were imported into Skyline and the performance of the peak picking was manually assessed for each peptide. Peptides, whose assignments failed our stringent evaluation criteria, were reanalyzed and reevaluated by MRM-MS. The quantitative data obtained by either SWATH-MS or MRM-MS were normalized for subsequent data analysis.

Because of the expected low abundance of certain peptides in some samples, the performance of the peak-picking algorithm was manually assessed for all peptides, and the peak boundaries were, if necessary, manually corrected to ensure faithful quantitative measurements. Only peptide assignments that fulfilled all of the following stringent evaluation criteria in at least one sample were considered for subsequent analyses: (i) the deviation between the observed and predicted retention time was less than 5 min, (ii) the dot product was greater than 0.8, (iii) the mass error was smaller than 5 ppm, and (iv) the maximal peak height exceeded 500 arbitrary intensity units. Using this strategy, quantitative measurements of 1095 HLA-B27-specific peptides were obtained across the eight different HLA-B27 subtypes and subsequently normalized to correct for differences in sample density and loading (Supplemental Table 1; Supplemental Table 3; Supplemental Fig. 3; Supplemental Fig. 4).

This comprehensive SWATH-MS dataset contained 92 peptides, which were recently quantified across all HLA-B27 allotypes by MRM-MS (18). To validate the accuracy of our SWATH-MS data, we compared the relative expression levels of these peptides to the published MRM-MS results (Supplemental Fig. 5A). Of note, 78 peptides (85%) showed an average variation of less than 20% across all HLA-B27 allotypes, confirming highly quantitative accuracy within the SWATH-MS dataset (Supplemental Fig. 5B). In addition, four peptides with both doubly and triply charged precursor ions were present in our SWATH-MS dataset. As expected, quantification of the differently charged precursor ions resulted in nearly identical relative expression levels across all HLA-B27 allotypes (Supplemental Fig. 5C).

In comparison to SWATH-MS, which results in MRM-like transition chromatograms, MRM-MS is a more focused and sensitive approach due to enhanced detection and use of target-optimized collision energies. Therefore, the 2211 peptides that did not pass our stringent SWATH-MS evaluation criteria were reanalyzed in 152 separate MRM-MS runs on a QTRAP^® 5500 mass spectrometer (SCIEX) operated in MRM mode and evaluated in Skyline using the same criteria as described for the SWATH-MS acquisitions. Pursuing this strategy, quantitative data of an additional 551 HLA-B27-specific peptides was obtained (Supplemental Table 3: Quantification results).

Taken together, we were able to quantify the abundance of 1646 peptides across the eight most frequent HLA-B27 allotypes using a combination of SWATH- and MRM-MS.

Cluster Analyses Confirm and Extend the Allele Specific-Binding Motifs of the Different HLA-B27 Allotypes

We applied cluster analysis to group peptides with similar relative binding patterns across the different HLA-B27 allotypes. Based on adjusted “figure of merit” graphs (Supplemental Fig. 6), which propose appropriate input parameters for the k-Means algorithm (45), we opted to group the 1646 peptides into 30 distinct clusters, which appeared to be sufficient to adequately separate the peptides according to their allotype-specific binding patterns (Fig. 2; Supplemental Fig. 7). The resulting clusters contained between 17 and 128 peptides, and we determined the peptide length distributions and the consensus-binding motifs for each cluster (Supplemental Fig. 8, Supplemental Fig. 9).

Fig. 2.

Cluster analysis. The 1646 quantified peptides were grouped into 30 distinct clusters according to their relative expression levels across the eight HLA-B27 allotypes. The diagrams show the relative expression level of each individual peptide within a cluster (thin gray lines). The thick red lines indicate the average expression level of all peptides within a cluster and the number of peptides within a cluster is shown below the cluster number.

The first eight cluster groups contained peptides that are mainly presented by a single HLA-B27 allotype (Fig. 2;Supplemental Fig. 7). As expected, the binding motifs of these clusters were similar to previously determined canonical consensus-binding motifs (4–18). However, the consensus-binding motifs detailed in these studies are defined by analyzing all peptides eluted from a specific HLA-B27 allotype. Conversely, peptides within any of the first eight cluster groups are predominantly presented by only one HLA-B27 allotype and therefore, must contain feature(s) that are favored by this allotype but less tolerated by the remaining subtypes. Indeed, a thorough comparison of these cluster-specific binding motifs revealed several allotype-favored peptide features for each HLA-B27 subtype and provided a means to attain the most stringent binding characteristics for each allotype (Table I).

Table I. Cluster properties.

The remaining 22 clusters are composed of peptides that are abundantly presented by more than one HLA-B27 allotype (Fig. 2; Supplemental Fig. 7). Of note, the consensus-binding motifs of these clusters show a mixture of some but not all allotype-favored peptide features of the respective HLA-B27 subtypes (Table I). For example, cluster 29 represents peptides that are highly presented by HLA-B*27:03, HLA-B*27:04, and HLA-B*27:06. Their consensus-binding motif shows a preferred usage of His, Arg, or Lys at P1 (HLA-B*27:03-favored); Leu at PΩ-2 (HLA-B*27:04-favored); and Gly, Ser, or Ala at PΩ-1 (HLA-B*27:06-favored). These observations suggest that the combination of allotype-specific features present in a given peptide dictates (at least to some extent) its abundance levels across the different HLA-B27 allotypes. Moreover, these features can provide a means to estimate the relative abundance level of a given peptide bound by each HLA-B27 allotype.

Nested Sets of Peptides Suggest Central Bulging as the Preferred Mechanism to Accommodate Extended Peptides

Approximately 6% of all HLA-B27-specific peptides are part of nested sets of peptides, which differ only in their N- and/or C-terminal extensions but share an identical core sequence (15, 18). Our dataset contained quantitative information of 55 of such nested sets (Supplemental Table 4). Recent studies suggested that central bulging—rather than terminal protrusion—is the preferred mechanism to accommodate extended peptides into the binding cleft of HLA-B27 allotypes (15, 18). In support of this theory, we observed that most peptides within one nested set show significant differences in their abundance levels across the HLA-B27 allotypes. More importantly, their abundance level distribution often correlates with the presence of HLA-B27 allotype-specific feature(s) if these peptides would anchor with their termini into the binding cleft. For example, the peptide VRMNVLADAL is predominantly presented by HLA-B*27:08 (presumably because of the Ala residue at PΩ-1). The extended peptide VRMNVLADALK is virtually absent from HLA-B*27:08 but mainly presented by HLA-B*27:05 due to the HLA-B*27:05-favored C-terminal Lys. These observations confirm and extend the hypothesis that most peptides anchor with their termini into the binding cleft.

Identification of Candidate Arthritogenic Peptides

To identify possible arthritogenic peptides, we screened our quantitative dataset for peptides that are presented in less amounts by HLA-B*27:06 and HLA-B*27:09 compared with disease-associated HLA-B27 subtypes. A total of 26 peptides fulfilled this criterion (Table II; Supplemental Table 5). The majority of these peptides carry a Tyr or Phe residue at their C terminus but do not contain other obvious similarities in addition to the conserved Arg at P2. To assess the degree by which these peptides are less presented by HLA-B*27:06 and HLA-B*27:09, we calculated for each peptide the quantitative difference to the disease-associated subtype with the lowest relative expression level (Supplemental Fig. 10). On average, this difference was 3.9% (±4.4%) with individual values varying between less than 1% to more than 20% (Supplemental Fig. 10). This reflects the complexity of the dataset and is consistent with our view that qualitative differences are rarely seen between HLA-B27 allotypic peptide ligands but rather there are subtle to quite substantial quantitative changes in peptide abundance. To highlight this, we performed an identical calculation across all 1646 quantified peptides and obtained an average quantitative difference of 6.7% (±7.5%) between the three lowest relative expression levels on different HLA B27 allotypes. Only two peptides considered as candidate arthritogenic peptides (ARYVFQSENTF and ARVLLVPDNTF) exceeded this error region in such that the quantitative difference between HLA-B*27:06 or HLA-B*27:09 to a disease-associated subtype was significantly greater than the average quantitative difference between the three lowest relative expression levels of a random peptide (Supplemental Fig. 10). However, the noise from allotypic binding preferences almost certainly masks the candidate peptides and even subtle quantitative differences are sufficient to cause a differential T cell response leading to the onset and progression of AS. Therefore, we believe each of these 26 peptides might represent the potential arthritogenic self-peptide and should be analyzed in future studies.

Table II. Putative arthritogenic peptides.

Peptide	Length	PΩ
GRVLLPEGGITA	12	A
HRAFSVFLF	9	F
SRVAEMFLF	9	F
TRYDLYHTF	9	F
ERLKDLFGKF	10	F
QRWDEAFRKF	10	F
ARVLLVPDNTF	11	F
ARYVFQSENTF	11	F
IRFPLMTIEEF	11	F
QRILDIVPPTF	11	F
SRWKIPSSW	9	W
TRGDLSSIRW	10	W
SRVLGVSDVITW	12	W
HRAPPIIGY	9	Y
IRLVPKLFY	9	Y
LRHPNILAY	9	Y
QRIYLVTTY	9	Y
QRLKIMEYY	9	Y
YRQWLKEEY	9	Y
FRFSGLLLGY	10	Y
MRTIATALEY	10	Y
SRFHLTPLSY	10	Y
ARTTIINEIQY	11	Y
ARVGSLSWNSY	11	Y
GRFGYLEGQEY	11	Y
KRFGFPEGSVELY	13	Y

DISCUSSION

This study describes for the first time the quantification of >1500 epitopes across several closely related HLA class I allotypes (HLA-B*27:02–HLA-B*27:09) using a combination of SWATH- and MRM-MS. SWATH-MS is a novel DIA method enabled on TripleTOF^® 5600⁺ mass spectrometers (SCIEX). Its potential lies in the ability to obtain quantitative MRM-like transitions for every peptide present in the sample and in the unique possibility to reinterrogate already acquired SWATH-MS data with a different set of peptides. However, due to the targeted nature of SWATH-MS, its data evaluation is strictly dependent on the availability of spectral and chromatographic information of all peptides of interest. This information is typically obtained by first analyzing the sample by standard DDA approaches in the presence of retention time alignment peptides (46). For this work, we generated such a spectral library in the open-source software Skyline (40) based on all HLA-B27 peptides identified in our previous study (18). Instead of spiking in peptides for retention time alignment, we exploited the presence of intrinsic peptides that were identified in each of the different HLA-B27 DDA acquisitions. Based on these peptides, we developed a custom-made “iRT-B27” retention time predictor (Supplemental Fig. 1) that allowed us to quantify 1095 peptides by SWATH-MS across all HLA-B27 allotypes.

MRM-MS - in comparison to SWATH-MS - is a more targeted approach and excels in unparalleled sensitivity but is limited by the number of peptides that can be analyzed in one run. We therefore opted to utilize MRM-MS to reanalyze only those peptides that were not identified by SWATH-MS. In addition, we did not measure the peptides of the iRT-B27 retention time predictor in each MRM-MS run but relied on a predetermined, MRM-gradient-specific regression curve of the iRT-B27 peptides for retention time alignment (correlation coefficient R² = 0.9776; Supplemental Fig. 2). Pursuing this strategy, we were able to analyze 2211 peptides across eight samples in only 152 individual MRM-MS runs resulting in quantitative measurements of additional 551 peptides, which constitute ∼25% of the peptides that escaped quantification by SWATH-MS.

Quantifying peptide ligands across various HLA molecules is hampered by the low abundance of some shared HLA-associated peptides and that not all peptides are expected to be present in each sample. Despite significant advances in MS instrumentation during the last decade, faithful quantitative measurements are not possible for peptides whose abundance levels are very low and/or only marginally exceed the sensitivity threshold of detection. In those cases, the analysis software might pick an incorrect peak or an unspecific signal in the background region, which inevitably leads to erroneous quantification measurements and the inability to unambiguously determine the presence (or absence) of a peptide in a given sample. To attenuate this issue, we manually assessed the performance of the peak-picking algorithm of all peptides and corrected the peak boundaries as necessary. Moreover, we applied strict evaluation criteria to our comprehensive datasets and annotated all peptides, which did not fulfill these criteria in some samples, accordingly in Supplemental Table 3.

Through the utility of the TM4:MeV software suite (43), we grouped all quantified peptides into 30 different clusters according to their relative binding to the eight HLA-B27 allotypes. Instead of accepting missing values in our cluster analysis, we opted to use all quantified values irrespective of whether a peptide assignment in a given sample passed our stringent evaluation criteria or not. This decision is based on the assumption that a peptide, which failed our evaluation criteria in a sample, is most likely either absent or of extremely low abundance. Either way, the quantified (background) signal is sufficiently low to not perturb the results of the cluster analysis.

The presence of clusters that contained peptides, which were exclusively presented by only one HLA-B27 allotype, enabled us to identify the most stringent binding characteristics for each HLA-B27 allotype (Table I). As expected, most of these allotype-favored peptide features are present in previously determined canonical consensus-binding motifs (4–18). However, not all features identified by canonical consensus-binding motifs represent allotype-favored peptide features, which highlights the importance of allotype-specific cluster analyses. Of note, allotype-favored peptide features also suggest the preferred HLA-B27 subtype of generally tolerated features. For example, all HLA-B27 allotypes are known to permit binding of peptides carrying a C-terminal Leu residue. However, it appears that HLA-B*27:06 is the most-suited subtype to accommodate this peptide feature.

The remaining 22 clusters, which contained peptides that are abundantly presented by more than one HLA-B27 allotype, showed a mixture of allotype-favored peptide features of the respective HLA-B27 subtypes (Table I). This observation suggests that (allotype-favored) peptide features represent a key element in determining the abundance levels of peptides across different HLA allotypes. However, not all individual peptides within these 22 clusters show appropriate peptide features implying that other factors such as (i) secondary structure formations, (ii) conformational changes, (iii) availability and prevalence of source proteins, and (iv) differential interactions with components of the peptide-loading complex act in concert to regulate peptide binding and abundance. Thus, it is not surprising that several individual peptides within these 22 clusters show unexpected primary sequences, which are not in agreement with the allotype-favored peptide features of the respective cluster. Nevertheless, this cluster analysis describes and highlights an easy, empirical, and universal approach to attain the most stringent binding characteristics for a given HLA molecule and bypasses the requirement for complex models and prediction algorithms, which often fail to produce unambiguous results. Unraveling the determinant selection of closely related HLA molecules will lead to a better understanding of a panel of clinically relevant topics such as allograft rejections, hypersensitivity reactions and other autoimmune diseases.

A detailed analysis of nested sets of N- and C-terminally extended peptides suggested that most peptides are able to anchor with their termini into the binding cleft of HLA-B27 subtypes resulting in central bulging rather than terminal protrusion. This notion, which also has been proposed previously (15, 18), is based on the observation that the abundance levels of peptides across the HLA-B27 subtypes only correlate with the presence of allotype-specific feature(s) if these peptides anchor with their termini into the binding cleft. While this is true for nearly all peptides that are extended by only one or two amino acids, some unusually long peptides (extensions of more than three amino acids resulting in peptides longer than 15 amino acids) appear to have similar abundance distribution patterns as their base peptide. This observation may suggest the usage of identical anchor residues and implies that the termini of these peptides stick out from the binding cleft resulting in terminal protrusions. This is particularly evident in case of the nested set surrounding the base peptide ARLALSPVPSH, which is mainly presented by HLA-B*27:05 (Supplemental Table 4). As expected, ARLALSPVPSHW, which carries a one amino acid C-terminal Trp extension, is predominantly presented by HLA-B*27:02. An additional Met extension at the C terminus (ARLALSPVPSHWM), which is a HLA-B*27:08-favored feature, completely abolishes presentation by HLA-B*27:02 in favor of HLA-B*27:08. All further C-terminally extended peptides (ARLALSPVPSHWMVA and ARLALSPVPSHWMVAL) show a nearly identical distribution pattern across the HLA-B27 allotypes as ARLALSPVPSHWM. Although additional experiments such as binding and refolding assays have to be carried out to support this notion, we speculate that the vast majority of peptides bind with their termini into the peptide-binding groove resulting in central bulging.

The main goal of this study was the identification of peptides that are less presented by HLA-B*27:06 and HLA-B*27:09 compared with disease-associated HLA-B27 subtypes. We identified 26 peptides fulfilling this criterion and therefore might encompass the sought-after arthritogenic peptide(s). In analyzing their characteristics, 16 peptides were found to carry either a Tyr or Trp residue at PΩ (Supplemental Table 4, Supplemental Fig. 10), which explains at least to some extent their low abundance in HLA-B*27:06 or HLA-B*27:09 eluates (7, 9, 14, 18, 47). Conversely, the amino acid sequences of the remaining 10 peptides show no obvious features that would imply an impaired binding to HLA-B*27:06 or HLA-B*27:09. This observation suggests that candidate arthritogenic peptides cannot be predicted solely based on primary sequences and consensus-binding motifs. In fact, direct quantitative analyses—as presented in this study—are necessary and indispensable for their identification.

In summary, we utilized a combination of targeted mass spectrometry techniques to identify 26 candidate arthritogenic peptides whose immunogenic potential will be analyzed in further immunological and structural studies.